This study proposes an RGB continuous wavelet transform (CWT) imaging method based on multi-channel time-series signals for tool wear prediction in milling machines. A regression model is constructed by integrating ResNet18 with transfer learning and a convolutional block attention module (CBAM) that incorporates both channel and spatial dimensions. First, two sets of sensor signals are transformed into RGB three-channel images using Continuous Wavelet Transform (CWT) and paired with the corresponding average tool wear labels to form the training and validation datasets, while the remaining sensor data are reserved for testing. After image resizing, tensorization and normalization using ImageNet statistical values, the images are fed into a lightweight ResNet18 backbone network. The CBAM is then applied to the final convolutional output to adaptively weight and enhance discriminative features. Model training employs the mean squared error (MSE) loss function, optimized using the Adam optimizer and combined with an early stopping strategy to prevent overfitting. The model demonstrates stable convergence on the validation set. Experimental results indicate that, compared with other methods, the proposed approach achieves superior performance metrics on the test set, thereby validating the effectiveness of CWT-based time–frequency imaging combined with ResNet and attention mechanisms for tool wear prediction.
Luo et al. (Fri,) studied this question.